GPT-5.6 Sol vs GPT-6 Astra
Humanity's Last Exam Humanity's Last Exam: Very hard expert questions across many fields, written to stump today’s models
GPT-5.6 Sol42.2
GPT-6 Astra52.7
Arena Text · overall Arena Text · overall: Human pairwise preferences in the stated text vote pool
GPT-5.6 Sol · xhigh1,485
GPT-6 Astra · max1,477
Arena Text · Chinese Arena Text · chinese: Human pairwise preferences in the stated text vote pool
GPT-5.6 Sol · xhigh1,526.6
GPT-6 Astra · max1,493
Arena Text · English Arena Text · english: Human pairwise preferences in the stated text vote pool
GPT-5.6 Sol · xhigh1,458
GPT-6 Astra · max1,447.9
Arena Text · French Arena Text · french: Human pairwise preferences in the stated text vote pool
GPT-5.6 Sol · xhigh1,474.5
GPT-6 Astra · max1,444.4
Arena Text · German Arena Text · german: Human pairwise preferences in the stated text vote pool
GPT-5.6 Sol · xhigh1,475.5
GPT-6 Astra · max1,457.1
Arena Text · instruction following Arena Text · instruction following: Human pairwise preferences in the stated text vote pool
GPT-5.6 Sol · xhigh1,482.3
GPT-6 Astra · max1,454.7
Arena Text · Japanese Arena Text · japanese: Human pairwise preferences in the stated text vote pool
GPT-5.6 Sol · xhigh1,470.9
GPT-6 Astra · max1,392.3
Arena Text · Korean Arena Text · korean: Human pairwise preferences in the stated text vote pool
GPT-5.6 Sol · xhigh1,442.1
GPT-6 Astra · max1,429.6
Arena Text · non-English Arena Text · non english: Human pairwise preferences in the stated text vote pool
GPT-5.6 Sol · xhigh1,450.8
GPT-6 Astra · max1,432.5
Arena Text · Polish Arena Text · polish: Human pairwise preferences in the stated text vote pool
GPT-5.6 Sol · xhigh1,461.4
GPT-6 Astra · max1,419.5
Arena Text · Russian Arena Text · russian: Human pairwise preferences in the stated text vote pool
GPT-5.6 Sol · xhigh1,470.4
GPT-6 Astra · max1,437
Arena Text · Spanish Arena Text · spanish: Human pairwise preferences in the stated text vote pool
GPT-5.6 Sol · xhigh1,440.6
GPT-6 Astra · max1,422.6
AA Omniscience Index AA Omniscience Index: Scores factual knowledge while penalizing incorrect answers to encourage admitting uncertainty
GPT-5.6 Sol19.4
GPT-6 Astra42.2
SimpleQA Verified v1 SimpleQA Verified v1: Accuracy on short factual questions with verified answers, answered without web access
GPT-5.6 Sol · max69.7
GPT-6 Astra75.8
CritPt (physics) CritPt (physics): Research-level physics problems that need many careful reasoning steps
GPT-5.6 Sol22.9
GPT-6 Astra29.1
AA Intelligence Index 4.3.2 AA Intelligence Index 4.3.2: Artificial Analysis combines tests of knowledge, reasoning, coding and tool use into one score
GPT-5.6 Sol39.2
GPT-6 Astra49.6
AA Omniscience · Accuracy AA Omniscience · Accuracy: Measures correct answers on the AA Omniscience factual knowledge test
GPT-5.6 Sol57.8
GPT-6 Astra60.6
AA Omniscience · Hallucination Rate AA Omniscience · Hallucination Rate: Measures how often attempted answers are incorrect on the AA Omniscience factual knowledge test · lower is better
GPT-5.6 Sol90.8
GPT-6 Astra46.5
AA-LCR (long context) 1.1 AA-LCR (long context) 1.1: Questions that need reading and connecting facts across very long documents (around 100K tokens)
GPT-5.6 Sol80.3
GPT-6 Astra79.7
GDP.pdf · AA all-pass GDP.pdf · AA all-pass: Models reason over professional PDF documents; an attempt passes only when every atomic criterion passes
GPT-5.6 Sol26.2
GPT-6 Astra30.4
MLCR-AA MLCR-AA: Models synthesize long medical records; the headline credits concise answers judged both accurate and complete
GPT-5.6 Sol15
GPT-6 Astra · max35
ARC-AGI-1 · public evaluation ARC-AGI-1 · public-eval: Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit
GPT-5.6 Sol93.9
GPT-6 Astra98.8
ARC-AGI-1 · semi-private ARC-AGI-1 · semi-private: Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit
GPT-5.6 Sol92.5
GPT-6 Astra97.5
ARC-AGI-2 · public evaluation ARC-AGI-2 · public-eval: Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit
GPT-5.6 Sol70.7
GPT-6 Astra96.7
ARC-AGI-2 · semi-private ARC-AGI-2 · semi-private: Models infer transformations from examples and solve unseen grid puzzles; the evaluation split is explicit
GPT-5.6 Sol67.1
GPT-6 Astra92.1
ARC-AGI-3 · semi-private ARC-AGI-3 · semi-private: Measures interactive game solving through relative human action efficiency
GPT-5.6 Sol1.1
GPT-6 Astra38.6
Chess Puzzles Chess Puzzles: Models choose the best next move from a novel chess position
GPT-5.6 Sol · max59
GPT-6 Astra · max74
FrontierMath Erdos FrontierMath Erdos: Models solve expert-written advanced mathematics problems on the named tier and edition
GPT-5.6 Sol · max0
GPT-6 Astra · max2.9
FrontierMath Tier 4 (v2) FrontierMath Tier 4 (v2): Models solve expert-written advanced mathematics problems on the named tier and edition
GPT-5.6 Sol · max82.9
GPT-6 Astra97.6
FrontierMath Tiers 1–3 (v2) FrontierMath Tiers 1–3 (v2): Models solve expert-written advanced mathematics problems on the named tier and edition
GPT-5.6 Sol · max89.1
GPT-6 Astra · max93.7
Mystery Game Puzzles Mystery Game Puzzles: Models choose the best move in puzzle positions from an undisclosed game
GPT-5.6 Sol · max58
GPT-6 Astra · max84
MathArena · ArXivMath · June 2026 MathArena · arxiv · 06/2026: Answer accuracy on the named mathematics competition under the recorded effort
GPT-5.6 Sol · max88.5
GPT-6 Astra · max94.4
MathArena · BrokenArXiv · June 2026 MathArena · arxiv_false · 06/2026: Answer accuracy on the named mathematics competition under the recorded effort
GPT-5.6 Sol · max67.3
GPT-6 Astra · max99.1
Vals MedCode Vals MedCode: Medical coding accuracy on patient encounters
GPT-5.6 Sol44
GPT-6 Astra48.5
Vals MedScribe Vals MedScribe: Accuracy of clinical notes produced from consultation transcripts
GPT-5.6 Sol85.2
GPT-6 Astra87.9
Vals Legal Research Bench Vals Legal Research Bench: Grounded legal research; a task passes only if all criteria pass
GPT-5.6 Sol48.1
GPT-6 Astra39.4
Vals Science Index Vals Science Index: Scientific problem solving across the index’s component evaluations
GPT-5.6 Sol41.5
GPT-6 Astra65.1
IFBench (instructions) IFBench (instructions): Checks whether the model follows precise formatting and content instructions
GPT-5.6 Sol69.6
GPT-6 Astra—
MathArena · ArXivMath · April 2026 MathArena · arxiv · 04/2026: Answer accuracy on the named mathematics competition under the recorded effort
GPT-5.6 Sol—
GPT-6 Astra · max90.8
MathArena · ArXivMath · August 2026 MathArena · arxiv · 08/2026: Answer accuracy on the named mathematics competition under the recorded effort
GPT-5.6 Sol—
GPT-6 Astra · max88.6
MathArena · ArXivMath · May 2026 MathArena · arxiv · 05/2026: Answer accuracy on the named mathematics competition under the recorded effort
GPT-5.6 Sol—
GPT-6 Astra · max95
MathArena · BrokenArXiv · April 2026 MathArena · arxiv_false · 04/2026: Answer accuracy on the named mathematics competition under the recorded effort
GPT-5.6 Sol—
GPT-6 Astra · max97.5
MathArena · BrokenArXiv · August 2026 MathArena · arxiv_false · 08/2026: Answer accuracy on the named mathematics competition under the recorded effort
GPT-5.6 Sol—
GPT-6 Astra · max81.9
MathArena · BrokenArXiv · May 2026 MathArena · arxiv_false · 05/2026: Answer accuracy on the named mathematics competition under the recorded effort
GPT-5.6 Sol—
GPT-6 Astra · max91
MathArena · ArXivLean · June 2026 MathArena · arxivlean · 06/2026: Answer accuracy on the named mathematics competition under the recorded effort
GPT-5.6 Sol—
GPT-6 Astra · max65.2
MathArena · ArXivLean · September 2026 MathArena · arxivlean · 09/2026: Answer accuracy on the named mathematics competition under the recorded effort
GPT-5.6 Sol—
GPT-6 Astra · max37.7
Vals LegalBench Vals LegalBench: Legal reasoning and legal-language understanding on LegalBench
GPT-5.6 Sol87
GPT-6 Astra—
| Benchmark | GPT-5.6 Sol | GPT-6 Astra |
|---|---|---|
| Humanity's Last Exam | 42.2 | 52.7 |
| Arena Text · overall | 1,485 · xhigh | 1,477 · max |
| Arena Text · Chinese | 1,526.6 · xhigh | 1,493 · max |
| Arena Text · English | 1,458 · xhigh | 1,447.9 · max |
| Arena Text · French | 1,474.5 · xhigh | 1,444.4 · max |
| Arena Text · German | 1,475.5 · xhigh | 1,457.1 · max |
| Arena Text · instruction following | 1,482.3 · xhigh | 1,454.7 · max |
| Arena Text · Japanese | 1,470.9 · xhigh | 1,392.3 · max |
| Arena Text · Korean | 1,442.1 · xhigh | 1,429.6 · max |
| Arena Text · non-English | 1,450.8 · xhigh | 1,432.5 · max |
| Arena Text · Polish | 1,461.4 · xhigh | 1,419.5 · max |
| Arena Text · Russian | 1,470.4 · xhigh | 1,437 · max |
| Arena Text · Spanish | 1,440.6 · xhigh | 1,422.6 · max |
| AA Omniscience Index | 19.4 | 42.2 |
| SimpleQA Verified v1 | 69.7 · max | 75.8 |
| CritPt (physics) | 22.9 | 29.1 |
| AA Intelligence Index 4.3.2 | 39.2 | 49.6 |
| AA Omniscience · Accuracy | 57.8 | 60.6 |
| AA Omniscience · Hallucination Rate | 90.8 | 46.5 |
| AA-LCR (long context) 1.1 | 80.3 | 79.7 |
| GDP.pdf · AA all-pass | 26.2 | 30.4 |
| MLCR-AA | 15 | 35 · max |
| ARC-AGI-1 · public evaluation | 93.9 | 98.8 |
| ARC-AGI-1 · semi-private | 92.5 | 97.5 |
| ARC-AGI-2 · public evaluation | 70.7 | 96.7 |
| ARC-AGI-2 · semi-private | 67.1 | 92.1 |
| ARC-AGI-3 · semi-private | 1.1 | 38.6 |
| Chess Puzzles | 59 · max | 74 · max |
| FrontierMath Erdos | 0 · max | 2.9 · max |
| FrontierMath Tier 4 (v2) | 82.9 · max | 97.6 |
| FrontierMath Tiers 1–3 (v2) | 89.1 · max | 93.7 · max |
| Mystery Game Puzzles | 58 · max | 84 · max |
| MathArena · ArXivMath · June 2026 | 88.5 · max | 94.4 · max |
| MathArena · BrokenArXiv · June 2026 | 67.3 · max | 99.1 · max |
| Vals MedCode | 44 | 48.5 |
| Vals MedScribe | 85.2 | 87.9 |
| Vals Legal Research Bench | 48.1 | 39.4 |
| Vals Science Index | 41.5 | 65.1 |
| IFBench (instructions) | 69.6 | — |
| MathArena · ArXivMath · April 2026 | — | 90.8 · max |
| MathArena · ArXivMath · August 2026 | — | 88.6 · max |
| MathArena · ArXivMath · May 2026 | — | 95 · max |
| MathArena · BrokenArXiv · April 2026 | — | 97.5 · max |
| MathArena · BrokenArXiv · August 2026 | — | 81.9 · max |
| MathArena · BrokenArXiv · May 2026 | — | 91 · max |
| MathArena · ArXivLean · June 2026 | — | 65.2 · max |
| MathArena · ArXivLean · September 2026 | — | 37.7 · max |
| Vals LegalBench | 87 | — |
Specs
| Spec | GPT-5.6 Sol | GPT-6 Astra |
|---|---|---|
| Input, per 1M | $4 | $10 |
| Output, per 1M | $20 | $50 |
| Cost per task? | $0.50 | $1.54 |
| Speed, tok/s | 85 | 59 |
| Context | 1.05M | 1.05M |
| Weights | Closed | Closed |
| Released | Jul 2026 | Sep 2026 |
| See images | ✓ | ✓ |
| Read PDFs | Partial | Partial |
| Hear audio | — | — |
| Watch video | — | — |
| Call tools | ✓ | ✓ |
| Force a tool call | Partial | Partial |
| Return strict JSON | ✓ | ✓ |
| Search the web | ? | ✓ |
| Run code | ? | ✓ |
| Use a computer | ? | ✓ |
What breaks when you switch
GPT-5.6 Sol → GPT-6 Astra
Monitoring can return HTTP 403›
Misalignment monitoring can return HTTP 403 before streaming or errors after output; match misalignment_policy_violation and handle stream errors.
No none reasoning effort›
The none reasoning effort is unsupported on GPT-6 Astra and GPT-6.1 Sol; use low instead.
Remove unsupported sampling parameters›
When reasoning effort is not none, remove temperature, top_p and top_logprobs, plus Chat Completions logprobs or Responses output logprobs inclusion.
GPT-6 Astra → GPT-5.6 Sol
Nothing known.